AWS prescriptive-guidance documentation change
Summary
Added new section 'Integrating AgentCore with CI/CD pipelines' detailing how AgentCore manages agent state, memory, and tools during deployment lifecycle. Included CodePipeline deployment steps.
Security assessment
The changes describe operational improvements for CI/CD pipelines but contain no evidence of addressing specific security vulnerabilities. Mentions of 'auditable access configuration' and 'compliance metrics' are general best practices, not security fixes.
Diff
diff --git a/prescriptive-guidance/latest/agentic-ai-serverless/cicd-and-automation.md b/prescriptive-guidance/latest/agentic-ai-serverless/cicd-and-automation.md index cc3cce8b2..4829f9a1c 100644 --- a//prescriptive-guidance/latest/agentic-ai-serverless/cicd-and-automation.md +++ b//prescriptive-guidance/latest/agentic-ai-serverless/cicd-and-automation.md @@ -5 +5 @@ -CI/CD capabilities in serverless AITypical CI/CD workflow for serverless AI projectsCI/CD for prompts and Amazon Bedrock agentsAWS services for CI/CD toolingSummary of CI/CD and automation +CI/CD capabilities in serverless AITypical CI/CD workflow for serverless AI projectsCI/CD for prompts and Amazon Bedrock agentsIntegrating AgentCore with CI/CD pipelinesAWS services for CI/CD toolingSummary of CI/CD and automation @@ -92,0 +93,30 @@ This approach prevents silent prompt degradation and ensures repeatable generati +## Integrating AgentCore with CI/CD pipelines + +Amazon Bedrock AgentCore extends traditional CI/CD automation by introducing a managed runtime and memory fabric for agent deployment, testing, and evolution. Current serverless pipelines automate the packaging and deployment of agent code (for example, through AWS CodePipeline, AWS CodeBuild, or AWS CDK). However, AgentCore integrates directly into this process to manage agent state, memory, and tool connectors as part of the deployment lifecycle. + +Key integration points of AgentCore with CI/CD pipelines are the following: + + * **Runtime registration and versioning** – Each deployed agent can be registered with AgentCore Runtime, which handles scaling, routing, and lifecycle orchestration. This approach replaces the need for maintaining custom registries or service discovery logic in CI/CD workflows. + + * **Memory snapshots and promotion** – During automated testing, AgentCore can persist agent memory snapshots, including learned context or state, and promote them alongside code artifacts through the pipeline. This capability enables context continuity between development, staging, and production environments. + + * **Tools configuration management** – Using AgentCore Gateway tools, teams can define integration points with other AWS services (for example, Amazon DynamoDB, Amazon S3, Amazon Bedrock FMs, or Amazon EventBridge) declaratively within the same pipeline. This configuration management capability helps provide consistent and auditable access configuration. + + * **Observability hooks for validation** – AgentCore exposes built-in telemetry for agent execution, enabling CI/CD pipelines to automatically validate performance, reasoning quality, and compliance metrics before deployment. + + + + +A CodePipeline deployment might consist of the following steps: + + 1. Build new agent code using CodeBuild. + + 2. Deploy the agent to AgentCore Runtime for execution. + + 3. Run automated integration tests that use AgentCore Memory to persist and compare state across runs. + + 4. Promote successful builds to production while updating AgentCore registries for discovery and orchestration. + + + + @@ -128,0 +159,2 @@ CI/CD is not just a best practice—it is a necessity for scaling safe and relia +With AgentCore integrated into CI/CD pipelines, agent deployment evolves from code delivery to continuous capability delivery. Reasoning, memory, and state become first-class deployable assets in modern serverless AI systems. +